'and' closing a list of items
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 15 of Qwen3-0.6B. Its strongest snippets are very consistent: they mostly peak on the same token. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron fires on the word 'and' when it joins the last two items in a list, such as 'Photography, Film, and Video' or 'sand, loam and clay'. Its negative side reacts to commas that start a new clause. Turning it up made the model drift into long, comma-heavy sentences.
Written by Claude from this neuron's activations and steering outputs. Confidence: medium. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
ography Visual Data: Photography, Film, and Video Using Documents as Data Qualitative Online Research: Using the Internet Observation and Med
-
is similar to use of arrowleaf balsamroot (pronghorn, bighorn sheep, and mule deer). Arrowleaf balsamroot is grazed year-round but
-
be flattered (if not shocked) that you'd be insightful enough to ask.⏎By Rod Machado<|endoftext|>PARIS — Should Palmyra be rebuilt? And under what conditions? No sooner had Syrian forces and the Russian Army freed the "
-
court in France. The book describes his court and the close relationships between the British and French royal families.⏎King in Exile: James II: Warrior, King and Saint by John Callow. Reassesses James's strategy for dealing with his
-
operation conditions, and inputs and outputs within and between each unit, vary in practice. These process variations can impact the key performance outcomes (titer, rate, and yield), which determine economic viability when the process is scaled up. The following descriptions
-
performed in the Globe Theatre. However, it was not only a period of prosperity, as there were a lot of religious conflicts (Catholics, Protestants and Puritans) with the resulting emigration to America of many Puritans
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: What if, if you don't do something, do not do something else. What if, if you don not do something. If you not do something, do not do something. If you not do something, do not do something,
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
exceed two feet. It’s rust-colored, with a long, slightly downwardly curved bill that’s two-toned — dark above, a dull orange below. The neck is rather long but thick. The back is
-
or CD-ROM. Includes the Catholic Encyclopedia, Church Fathers, Summa, Bible and more all for only $19.99...⏎A Spanish historian; born at Cuellar, in the
-
= 871) from 5 schools attended one of 2 DST performances. Participants demonstrated increased knowledge on 5 fact-based questions (mean improvement range, 19%- 35%; p < .001 for all
-
night to make it to Willamette’s 1929 freshman orientation. A month later the stock market collapsed — his bank failed, his savings account was wiped out and his scholarship was withdrawn for lack of funds.
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L13:1632 suppresses words inside headlines and titles
- L13:1180 suppresses
- L12:2605 excites
- L14:1628 suppresses
- L13:1026 suppresses
- L11:559 excites
- L5:2313 excites
- L10:1459 suppresses
- L8:1473 suppresses
- L10:371 excites
- L11:45 suppresses
- L14:2657 excites
- L11:840 weakly excites
- L9:2357 weakly excites
- L14:2899 weakly suppresses
- L13:1610 weakly excites
- L10:1794 weakly excites
- L12:3061 weakly suppresses
- L11:514 weakly excites
- L11:237 weakly excites
- L12:1959 weakly suppresses
- L10:1643 weakly suppresses
- L6:676 weakly suppresses
- L11:1968 weakly suppresses
Feeds later layers
- L16:2577 strongly excites
- L17:2221 strongly excites
- L16:1145 excites commas inside short lists
- L20:427 suppresses
- L16:650 excites
- L20:373 suppresses
- L16:1463 suppresses
- L21:739 excites 'both X and Y' pairing words
- L18:1327 suppresses
- L16:2432 excites
- L19:1414 suppresses
- L26:347 excites the 're' in self-reliant and similar
- L24:205 excites
- L17:2234 suppresses
- L19:1245 excites comma opening a list of parallel items
- L19:1262 excites
- L26:236 suppresses
- L18:1498 suppresses
- L26:65 suppresses
- L19:2451 suppresses
- L18:862 suppresses hedges like 'it turns out' and 'if any'
- L16:1669 suppresses
- L17:572 suppresses
- L17:2610 excites and or comma closing a list
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- and
- 和完善
- and
- anden
- 时间和
- 性和
- enschaft
- && ⏎
- ()'
- 清代
Suppresses
- ;
- ,“
- ;
- 其它
- ;}⏎⏎
- ;]/
- ,
- (Connection
- emies
- ٬